Multi-function radar work mode recognition based on residual shrinkage reconstruction recurrent neural network

In modern electronic warfare, multi-function radar work mode recognition is increasingly crucial. However, the challenges posed by complex electromagnetic environments, such as lost pulses, spurious pulses, and measurement errors, along with the reliance of traditional multi-task learning strategies on clean samples, make it difficult for existing algorithms to achieve satisfactory recognition performance in real-world scenarios. To address these issues, this paper introduces a novel residual shrinkage reconstruction recurrent neural network (RS-RRNN). The network uses a Gated Recurrent Unit as its backbone to extract temporal features and enhances feature extraction by reconstructing the GRU's input, while also reducing dependence on clean samples. These features are then processed through a residual shrinkage structure to reduce noise, which significantly improves the model's robustness in non-ideal scenarios. Simulations demonstrate that RS-RNN has better performances in accuracy and robustness than existing networks.